4-layer 67M-parameter model

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

• Layer-wise geometry summaries (Sbmax, AUSN) predict internal few-shot thresholds θ50claim0.715
  Claim that geometry-to-behavior correlates exist
• 2/3 Layer of LLMconcept0.715
  The layer approximately two-thirds through an LLM's transformer stack, reported to best predict human brain activity; identified as promising for consciousness indicators.
• deeper layers (16–28)concept0.709
  Layers where anchoring weakens systematically due to representational drift.
• Peak layer ℓ* median 10, IQR 0.384finding0.699
  Median layer where S(ℓ) peaks, across seeds.
• 1:50 physical modelmethod0.690
  A 1:50 scale model used for overall design simulation of the Athens Megaron spaces and floors.
• Systematic layer 20-28 degradation in S(ℓ) to S ≈ −2.40 by layer 27 on LLaMAfinding0.689
  Validates representational drift theory: later layers specialize for next-token prediction, increasing dr
• Decomposition of all 24 weight matrices in a 67M-parameter LM yields ~10,000 parameter subcomponentsfinding0.687
  Quantitative result of VPD application; the network's 24 matrices decompose into approximately 10,000 rank-one subcomponents.
• Mid-Layer Emotion Representation Peakconcept0.685
  Empirical observation that steering efficiency peaks at middle transformer layers, consistent with emotion representation literature